Experimental Evaluation of Textile Cloth Defects based on Color and Texture Analysis by using Moderated Deep Learning Algorithm

Textile defect detection plays a crucial role in quality control within the textile manufacturing industry, as undetected defects can lead to significant financial loss and a negative reputation. Traditional methods for defect detection, such as manual inspection and basic machine learning models, often fail to capture complex patterns, resulting in suboptimal accuracy. This study proposes a novel hybrid model integrating EfficientNet for fine-grained feature extraction, GRU (Gated Recurrent Unit) for sequence-based defect detection, and DenseNet-LightGBM for robust classification. The model was evaluated on a dataset containing various textile defects such as missing threads, color blotches, and texture inconsistencies. Compared with nine existing models, including ResNet, VGG16, AlexNet, and InceptionV3, the proposed model achieved a superior accuracy of 94.75%, outperforming the other models in all key performance metrics. Furthermore, the model demonstrated strong precision (95.35%), recall (94.10%), and F1-score (94.72%), confirming its robustness in defect detection. The results indicate that this hybrid deep learning approach is not only more accurate but also more efficient for real-time textile defect detection.